Building AI agents

Technology that is sufficiently advanced is indistinguishable from magic.

It was classic — advanced technology feels like magic — moment for me when Jiquan Ngiam showed me the demo of AI agents he was building. AI agents are all the rage in gen AI but few of them actually work. I have not seen agents that matched my high expectations, but then I was on a call for the demo of Lutra.Ai.

In the demo couple of things became very clear to me

• AI agents when they work will look like magic

• It is really hard to make them work with low error rate

• General AI agents are 10x harder than specific agents

I could tell in the demo that Ngiam is a 10x engineer. His impressive background is a proof for it. He did his Ph.D at Stanford under the iconic Andrew Ng. He helped Andrew with his world famous machine learning course and later became engineer #1 at Coursera and is currently building Lutra.AI and ai agent for knowledge workers.

Since the last one month I have been interacting with Lutra, the product has been evolving rapidly and Ngiam and team are clearly rockstar executioners and trying out new formats of building AI agents that I have not seen before.

So its natural for me to have Ngiam on the podcast and talk about AI agents and more.

We talked about:

• being first engineer at course era

• working with deep learning founder Andrew Ng

• building agentic ai application for knowledge workers

• scale of future models and of course AGI

• working at google brain and the rise of transformers

& a whole lot more.

Check out my full conversation with the Founder of Lutra.AI, Engineer #1 at Coursera — Jiquan Ngiam on YouTube, Spotify or wherever you consume audio.

If you are an investor who is actively investing in early stage companies, don’t forget to checkout my latest product checksiwrote.com.

Nataraj

Follow me on LinkedIn & Twitter